We just got a few new large memory machines with Haswell and Broadwell 
architectures. 

I'm wondering what is the current recommendation for getting Julia to run 
as fast as possible on these machines.

We have a license for the Intel Compilers and MKL.

So the question is: What's the best way to build from source? 

- Intel compilers?
- GNU compilers? 
- MKL or no MLK?
- Setting for MARCH? 
- Setting for JULIA_CPU_TARGET? Is native the best choice here?

And next, what is the best way to benchmark and test the resulting compile?

It looks like make testall does the testing, but is there a standard 
benchmark command to confirm we're getting the "fastest" julia? 

I did look around for help in the forms and via google, and found bits and 
pieces. (e.g. JULIA_CPU_TARGET, for instance...). Just wondering what the 
latest recommendations and experiences are!

Thanks!

David Gleich

Reply via email to